I have data points for 180 days collected at an interval of 15 minutes. But some of the data points are missing. So ideally there should have been 180*24*4 data points but I am having less. If I want to use this data from time series forecasting, what's the best method to handle these missing values?
The easiest method that comes to my mind is to remove the days with less than the required number of data points (24*4). This will preserve the seasonality per day but I am afraid it will destroy the seasonality per week. Is synthesizing the data point from the average better?
Read Box, Jenkins book on time series analysis. It has a section for handeling missing values as I remember. (I think many ARIMA libraries there handle missing values)
If you use XGBoost the library handle missing values in default.
Hi Alex Mathew
You may want to use the following software package for analyzing time series with gaps and/or missing values:
https://geodesy.noaa.gov/gps-toolbox/JUST.htm
The software is in both MATLAB and Python. It has several tools for time series analysis, such as spectral analysis, wavelet analysis, breakpoint detection, forecasting, near-real-time monitoring, etc. Please let me know if you had any questions
Best regards,
Ebrahim
Hi,
there is good explanation to prepared of time series data from website.
https://machinelearningmastery.com/how-to-load-visualize-and-explore-a-complex-multivariate-multistep-time-series-forecasting-dataset/
- Badges
- Science topic